I’m working on a small embedded Rust project based on the https://github.com/Rahix/avr-hal-template.git for AVR cpus.

I want to add some unit tests, so I removed the test = false line from the Cargo.toml generated by the template. I’ll add my own test runner with harness = false, and got a working version already, but I don’t understand one step I had to take to get things working.

When I run cargo test immediately after removing test = false I get a bunch of errors like this (for several crates besides ufmt_write):

error[E0152]: duplicate lang item in crate `core` (which `rustc_std_workspace_core` depends on): `sized`.
  |
  = note: the lang item is first defined in crate `core` (which `ufmt_write` depends on)
  = note: first definition in `core` loaded from /home/niels/git/capevm-rust/capevm/target/avr-atmega128/debug/deps/libcore-0538a43361b060d2.rmeta
  = note: second definition in `core` loaded from /home/niels/git/capevm-rust/capevm/target/avr-atmega128/debug/deps/libcore-f17641206fc9a410.rmeta

I can fix this by removing the panic = "abort" line from my [profile.dev] in Cargo.toml, and adding "panic-strategy": "abort" to my target specification file (avr-atmega128.json in my case).

Why is this necessary?

Copilot tells me the conflicting core definitions can be caused when code compiled for the unwind and abort strategies are mixed, which somewhat makes sense to me, but I’d like to better understand why it makes a difference whether I set it to "abort" in Cargo.toml or in the target spec file, and why this isn’t necessary for a normal (non-test) build?

What’s especially puzzling me is that removing the panic = "abort" line from in Cargo.toml is necessary. With the line there, I still get the same error, but it compiles fine after removing it.